Skip to content
96Astro API

Blog

Face Reading API: responsible AI face reading for apps

· 96Astro API team · 7 min read · Photo readings, Guides

An anonymous sculptural face, camera and consent shield illustrate responsible photo reading.

Build an AI face reading (physiognomy) feature: 24 endpoints for features and life readings, outline tracing, expression scores, consent and the rules that keep it safe.

In brief

  • A face reading API reads one photo of a face in the tradition of physiognomy (samudrika shastra in India, mian xiang in China) and returns readings of the features and of life areas.
  • The 96Astro face-reading family has 24 capabilities: twelve feature reads, ten life-area readings, a whole reading in one call and a traced outline with expression scores.
  • Consent is required, and model instructions forbid identification and sensitive-trait inference. Product validation must enforce these restrictions; provider retention must also be reviewed.
  • It is built for reflection and entertainment. It must never be used to decide anything about a person.

Face reading apps go viral because the input is a selfie: no birth time, no place search, one tap. That ease is also the risk. A face is sensitive personal data, and a careless reading can hurt people. This guide shows how to build an AI face reading feature that is engaging and responsible, using a face reading API.

What face reading (physiognomy) is, and what an API adds

Physiognomy traditionally claims to read character and tendencies from the shape and proportions of the face: the forehead, eyes, nose, mouth, chin and ears. Many cultures have a tradition of it, from Indian samudrika shastra to Chinese face reading. Historically it was done by a practitioner looking at a person. Facial appearance does not reliably establish personality, trustworthiness, health or future events; present traditional narratives as entertainment rather than factual assessments.

A face reading API replaces the practitioner's eye with a vision model that has a narrow brief. It reports what it can see in the photo, then reads those observations in the traditional framework. Every section comes back in the same three parts:

FieldMeaning
observedWhat the model saw in this photo
readingThe physiognomy reading of those observations
adviceOne practical suggestion

Because the observations are returned, your users can see the basis of every reading. If the ears are hidden by hair, the output should mark that observation unavailable. Check this behaviour on test photos; models can still hallucinate visible details.

The 24 face reading endpoints

GroupCapabilities
Featuresface-shape, face-eyes, face-nose, face-cheeks, face-mouth, face-chin-jaw, face-ears, face-skin, face-symmetry, face-proportions, face-facial-hair, face-outline
Life areasface-personality, face-career, face-wealth, face-love, face-marriage, face-family, face-health, face-wisdom, face-luck, face-challenges
Everythingface-reading, every section for one model call
Drawingface-outline-image, the face and its features as shapes, with expression scores

face-health is a reading of energy and everyday wellbeing in the tradition. It must not diagnose or predict illness, and face-skin comments on clarity as the photo shows it, never on skin colour.

A first face reading request

curl https://api.96astroapi.com/v1/face-reading/face-reading \
  -H "Authorization: Bearer $ASTRO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "image_base64": "<base64 of the JPEG>",
    "mime_type": "image/jpeg",
    "consent": true,
    "language": "en",
    "tier": "fast"
  }'

The fields are the same as for palm reading: a PNG, JPEG or WebP image up to 3.75 MiB decoded, consent: true, an optional language and a tier (fast or balanced). A photo the model cannot use returns 422 image_not_usable and is not charged. Full details are in Palm and face readings.

Tracing the face and reading the expression

face-outline-image returns the outline of the face and its features as points on a 0 to 1000 grid, with an overlay_svg you lay over your own copy of the photo. It also scores how the face looks in that photo:

ScoreMeaning
energyHow lively the face looks
calmnessHow relaxed it looks
happinessHow much it smiles
approachabilityHow open it looks

These are subjective model outputs, not measurements of emotion, mental state or character. Prefer observable descriptions such as a visible smile; do not present calmness, energy or approachability as established facts about the person.

Designing a face reading app screen

A flow that works:

  1. Photo guidance. Face the camera straight on at eye level; the whole face in view with hair off the forehead; bright, even light from the front; a neutral expression with eyes open; no sunglasses, hats or heavy beauty filters; one face in the photo.
  2. Consent. A clear checkbox: the person in the photo agreed, and is an adult or has a guardian's agreement.
  3. Trace. Show the outline over the photo with a before-and-after slider and the four expression scores.
  4. Features. Twelve cards, run one by one or all at once.
  5. Life readings. Ten cards: personality, career, wealth, love, marriage, family, vitality, wisdom, luck and challenges.
  6. History. Save results on the device so reopening a photo costs nothing.

The 96Astro dashboard has this exact flow under Face reading, so you can try it with a sandbox key before you write a line of code.

Responsible face reading: the rules we build in

Face analysis has a bad history, and it is right to be careful. The model behind our face reading API is instructed to:

  • never identify anyone, or say who someone looks like;
  • never infer race, religion, caste, sexuality, disability or age;
  • never comment on skin colour;
  • never diagnose or predict illness.

Your product must hold the same line:

  • No decisions about people. Face readings must not be used for hiring, lending, insurance, tenancy, dating decisions about a third party, or any judgement about a person.
  • Consent from the person in the photo. Not from whoever uploads it. Do not allow photos of children without a guardian's agreement.
  • Say what it is. "An AI reading of your photo, in the tradition of face reading, for reflection."
  • Check your local law. In many countries a face photo is biometric or personal data with specific rules.

These rules are also good product design. A face reading app that people trust gets shared; one that feels creepy gets uninstalled.

Face reading costs, and how to keep them low

Face readings are metered by model cost, like other AI readings, at the vision tier you choose. To keep costs low:

  • Use face-reading when you show many sections. It is one model call for every section.
  • Cache by photo. Reuse a saved result within the same consent scope, language and model version; a new model call can produce different wording or scores.
  • Check the photo first with face-outline-image before a user buys a bundle of readings.
  • Use the fast tier unless a premium product justifies the balanced one.

Face reading feature ideas

  • A selfie reading as the no-friction entry point to an astrology app, before asking for birth details.
  • "Your face and your chart." Show the face personality reading next to the Moon sign and ascendant from a birth chart.
  • A printable face reading report as a paid add-on, made from the reading you already have, with no photo sent again. See jobs and PDFs.
  • A daily expression check-in using the four expression scores, framed as "how you look today", not "who you are".

Face reading vs palm reading

Face readingPalm reading
InputOne face photoOne palm photo
FeelsInstant, social, shareablePersonal, traditional
SensitivityHigher: faces are identifyingLower, but still personal data
Couple featureNoYes, palm-compatibility
DrawingOutline with expression scoresFour main lines with confidence

Many apps offer both: face reading as the quick hook, palm reading as the deeper paid reading. Read the palm reading API guide for the palm side.

Frequently asked questions

Is the face photo stored?

The application does not intentionally retain the photo as a report asset. It forwards the image to a vision provider, so review that provider's retention terms and your own logging configuration too. Only keep your own copy if your user agreed, and let them delete it.

Can the face reading API recognise who someone is?

No. The model is told never to identify a person or say who they resemble. It reads features in the physiognomy tradition only.

Does it judge attractiveness, ethnicity or age?

The model is instructed not to infer race, religion, caste, sexuality, disability or age, and never comments on skin colour. There is no attractiveness score.

What are the expression scores for?

They are model-generated impressions, not validated emotion or character measurements. Do not use them to rank people or claim knowledge of their internal state.

Can I use face reading in a dating app?

Only as an opt-in reading of a user's own photo for themselves. It must never be used to rank, filter or judge other people.

Which languages are supported?

All 26 languages of the API, including Hindi, Marathi, Bengali, Tamil, Telugu, Kannada and Malayalam, and major European and Asian languages.

Get started

Open the Face Reading API page for request examples, read the Palm and face readings guide for every field and limit, and try the Face reading screen in your dashboard with a free sandbox key.

Get the next guide by email

One email a month with new guides and API changes. No account needed.

About once a month: new guides, API features and astrology-developer news. We email a confirmation link first; unsubscribe with one click from any newsletter. See our privacy policy.

Send us feedback

What is it about?